United States Macroeconomic Data Analysis using Python¶

This dataset contains month-wise details about all the macro-economic factors of US over two decades from 05/2002 to 05/2022 which can be used to determine how different factors affect a particular factor.

Sourced and compiled data from official websites of US government like FRED, CENSUS, OECD, Conference Board.

Collected economic data like Income, GDP, different indices, mortgage etc. from FRED official website, demographic data from OECD and consumer confidence Index from Conference Board.

Data obtained from : https://www.kaggle.com/datasets/sagarvarandekar/macroeconomic-factors-affecting-us-housing-prices/data

We will use this data to perform EDA and develop some predictive models and measure their performance.

Data Dictionary¶

  • UNRATE(%) = Unemployment Rate
  • CONSUMER CONF INDEX = Consumer Confidence Index provided by Conference Board
  • PPI-CONST MAT. = Producers Purchase Index - Construction Materials
  • CPIALLITEMS = Consumer Price Index - All Items for the US
  • INFLATION(%) = Inflation rate in the US
  • MORTGAGE INT. MONTHLY AVG(%) = Average Mortgage interest rate of all Weeks of a particular month
  • MED HOUSEHOLD INCOME = Median Household Income in the US
  • CORP. BOND YIELD(%) = Corporate Bond Yield (%), the percentage Yield of corporate Bonds
  • MONTHLY HOME SUPPLY = Indicates how long the current homes would last given no new homes
  • % SHARE OF WORKING POPULATION = % Share of working population between the age 18 and 60
  • GDP PER CAPITA = GDP per capita
  • QUARTERLY REAL GDP = Quarterly Real GDP
  • QUARTERLY GDP GROWTH RATE (%) = Quarterly GDP growth rate (%)
  • CSUSHPISA = S&P/Case-Shiller U.S. National Home Price Index given by FRED

Data Analysis¶

Let us start by importing the libraries we are going to use in our analysis.

In [1]:
import pandas as pd
import plotly.express as px
import numpy as np
import plotly.graph_objects as go
from plotly.subplots import make_subplots
import plotly.express as px
import seaborn as sns
from sklearn.linear_model import LinearRegression
In [2]:
df = pd.read_csv('US_Macroecon_data.csv')
In [3]:
df.describe()
Out[3]:
UNRATE(%) CONSUMER CONF INDEX PPI-CONST MAT. CPIALLITEMS INFLATION(%) MORTGAGE INT. MONTHLY AVG(%) MED HOUSEHOLD INCOME CORP. BOND YIELD(%) MONTHLY HOME SUPPLY % SHARE OF WORKING POPULATION GDP PER CAPITA QUARTERLY REAL GDP QUARTERLY GDP GROWTH RATE (%) CSUSHPISA
count 241.000000 241.000000 241.000000 241.000000 241.000000 241.000000 224.000000 241.000000 241.000000 236.000000 241.000000 241.000000 241.000000 241.000000
mean 6.074689 90.809544 206.949863 95.539665 2.296497 4.697956 53273.982143 4.471162 5.974274 66.410104 52896.082988 16536.012095 0.490060 175.306996
std 1.987058 25.871004 40.479900 11.087025 1.641645 1.119850 7475.321506 1.079004 1.895763 0.802918 8840.592318 1708.435162 1.453910 36.780758
min 3.500000 25.000000 143.800000 75.859538 -2.097161 2.684000 42409.000000 2.140000 3.300000 64.924129 37860.000000 13477.356000 -8.937251 120.724000
25% 4.700000 70.400000 183.300000 87.722400 1.463784 3.802500 49007.250000 3.690000 4.600000 65.623251 46977.000000 15304.517000 0.293599 147.395000
50% 5.600000 94.500000 206.200000 96.819215 2.070508 4.457500 50303.000000 4.340000 5.500000 66.739214 51554.000000 16253.726000 0.580001 169.812000
75% 7.300000 108.200000 221.700000 103.255463 2.969762 5.812500 59039.000000 5.410000 6.700000 67.128435 58745.000000 17896.623000 0.833911 189.707000
max 14.700000 138.400000 352.857000 123.322800 8.581511 6.806000 68703.000000 6.750000 12.200000 67.298433 74737.000000 19806.290000 7.547535 304.831000
In [4]:
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 241 entries, 0 to 240
Data columns (total 15 columns):
 #   Column                         Non-Null Count  Dtype  
---  ------                         --------------  -----  
 0   DATE                           241 non-null    object 
 1   UNRATE(%)                      241 non-null    float64
 2   CONSUMER CONF INDEX            241 non-null    float64
 3   PPI-CONST MAT.                 241 non-null    float64
 4   CPIALLITEMS                    241 non-null    float64
 5   INFLATION(%)                   241 non-null    float64
 6   MORTGAGE INT. MONTHLY AVG(%)   241 non-null    float64
 7   MED HOUSEHOLD INCOME           224 non-null    float64
 8   CORP. BOND YIELD(%)            241 non-null    float64
 9   MONTHLY HOME SUPPLY            241 non-null    float64
 10  % SHARE OF WORKING POPULATION  236 non-null    float64
 11  GDP PER CAPITA                 241 non-null    int64  
 12  QUARTERLY REAL GDP             241 non-null    float64
 13  QUARTERLY GDP GROWTH RATE (%)  241 non-null    float64
 14  CSUSHPISA                      241 non-null    float64
dtypes: float64(13), int64(1), object(1)
memory usage: 28.4+ KB
In [4]:
#df['DATE'] = pd.to_datetime(df['DATE'])
In [6]:
#df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 241 entries, 0 to 240
Data columns (total 15 columns):
 #   Column                         Non-Null Count  Dtype         
---  ------                         --------------  -----         
 0   DATE                           241 non-null    datetime64[ns]
 1   UNRATE(%)                      241 non-null    float64       
 2   CONSUMER CONF INDEX            241 non-null    float64       
 3   PPI-CONST MAT.                 241 non-null    float64       
 4   CPIALLITEMS                    241 non-null    float64       
 5   INFLATION(%)                   241 non-null    float64       
 6   MORTGAGE INT. MONTHLY AVG(%)   241 non-null    float64       
 7   MED HOUSEHOLD INCOME           224 non-null    float64       
 8   CORP. BOND YIELD(%)            241 non-null    float64       
 9   MONTHLY HOME SUPPLY            241 non-null    float64       
 10  % SHARE OF WORKING POPULATION  236 non-null    float64       
 11  GDP PER CAPITA                 241 non-null    int64         
 12  QUARTERLY REAL GDP             241 non-null    float64       
 13  QUARTERLY GDP GROWTH RATE (%)  241 non-null    float64       
 14  CSUSHPISA                      241 non-null    float64       
dtypes: datetime64[ns](1), float64(13), int64(1)
memory usage: 28.4 KB
In [5]:
import matplotlib
import matplotlib.pyplot as plt
import matplotlib.dates as dates
%matplotlib inline


datearray = df['DATE'].apply(lambda x: np.datetime64(x[6:]+'-'+x[3:5]))

i=1
# we iterate through all values and graph them
for column in df.columns:
    # there's no sense in graphing dates to dates
    if column == 'DATE':
        continue
        
    print("FIG."+str(i)+" "+column+" over time.")
    i += 1
    
    # defining x and y axis
    x_axis = (datearray)
    y_axis = df[column]

    # plotting the figure
    plt.rcParams["figure.figsize"] = (20,8)
    plt.xlabel("dates")
    plt.ylabel(column)
    ax = plt.subplot()
    ax.stem( x_axis, y_axis)
    ax.xaxis_date()
    plt.show()

FIG.1 UNRATE(%) over time.

FIG.2 CONSUMER CONF INDEX over time.

FIG.3 PPI-CONST MAT. over time.

FIG.4 CPIALLITEMS over time.

FIG.5 INFLATION(%) over time.

FIG.6 MORTGAGE INT. MONTHLY AVG(%) over time.

FIG.7 MED HOUSEHOLD INCOME over time.

FIG.8 CORP. BOND YIELD(%) over time.

FIG.9 MONTHLY HOME SUPPLY over time.

FIG.10 % SHARE OF WORKING POPULATION over time.

FIG.11 GDP PER CAPITA over time.

FIG.12 QUARTERLY REAL GDP over time.

FIG.13 QUARTERLY GDP GROWTH RATE (%) over time.

FIG.14 CSUSHPISA over time.

We observe several notable trends in our data:

  • Changes in the inflation rate and mortgage interest rate appear to mirror each other, indicating a strong positive correlation.
  • The unemployment rate experienced a significant spike in 2020, likely due to the COVID-19 pandemic.
  • Both the inflation rate and mortgage interest rate showed substantial increases as the unemployment rate began to decline following its 2020 peak, suggesting a relationship among these three variables.
  • The 2008 housing crisis impacted employment trends differently compared to the COVID-19 pandemic. While the 2020 spike was abrupt and volatile, the 2008 crisis exhibited a more gradual exponential rise.
  • Both inflation and GDP growth rates displayed marked volatility in response to the economic shocks of 2008 and 2020.

Let us view the attributes plotted next to each other.

In [8]:
df = pd.read_csv('US_Macroecon_data.csv')
df['DATE'] = pd.to_datetime(df['DATE'])

Ndf = pd.DataFrame()
for column in df.columns:
    if df[column].dtype == 'int64' or df[column].dtype == 'float64':
        Ndf[column] = df[column]
Ndf['DATE'] = df['DATE']
Ndf['NDATE'] = Ndf['DATE'].dt.strftime('%m-%Y')
Ndf = Ndf.groupby(by=['NDATE']).mean().reset_index()

nrows = 7 
ncols = 2

cols = Ndf.columns[1:]
                                                     
fig = make_subplots(rows=nrows, cols=ncols,subplot_titles=cols)

for index,col in enumerate(cols):
    fig.add_trace(go.Scatter(x=Ndf['NDATE'], y=Ndf[col],mode='markers', 
                            hovertemplate="<br>".join(["Date: %{x}",
                                                       "Value: %{y}",
                                                       ])),      
                  row=index//ncols+1, col=(index%ncols)+1)
    
fig.update_layout(height=1700, title_text="Metrics over Time",showlegend=False)
fig.update_xaxes(tickangle=45)

fig.show()

Over the past two decades, CPIALLITEMS, median household income, GDP per capita, and quarterly real GDP have shown a steady upward trend. Conversely, the percentage share of the working population, corporate bond yields, the monthly average mortgage interest rate, and CSUSHPISA have generally been on a downward trajectory. Meanwhile, quarterly GDP growth rate, inflation, and consumer confidence have exhibited fluctuations over the same period.

Let us now take a look at the measure of volatility between the different metrics. Some measures change rapidly while others are more gradual.

In [9]:
std = pd.Series(data=[Ndf[col].std()/Ndf[col].mean() for col in Ndf.columns[1:]],index=Ndf.columns[1:])
std = std.apply(lambda x : round(x,3))

fig = go.Figure(data=[go.Table(
    header=dict(values=['Metric','Volatility'],
                fill_color='paleturquoise',
                align='left'),
    cells=dict(values=[std.index,std.values],
               fill_color='lavender',
               align='left'))
])
fig.update_layout(title='Levels of Volatility')
fig.show()

The quartrly GDP growth rate has the highest level of volatility and inflation being a somewhat close second. Quarterly GDP is differentiated as it's a rate and not a measurement like the other metrics.

Let us plot Inflation and Quarterly GDP to see how they're changing

In [10]:
pct_change = Ndf.copy()
pct_change.set_index('NDATE',inplace=True)
pct_change = pct_change.pct_change()
for col in pct_change.columns:
    newcol = col+' pct_change'
    pct_change = pct_change.rename(columns={col:col+' pct_change'})
    pct_change[newcol] = pct_change[newcol].map('{:.0%}'.format)
    
combined = Ndf.join(pct_change,how='left',on='NDATE')

nrows = 2 
ncols = 1

cols = ['INFLATION(%)','QUARTERLY GDP GROWTH RATE (%)']
                                                     
fig = make_subplots(rows=nrows, cols=ncols,subplot_titles=cols)

position = 0
for index,col in enumerate(cols):
    
    row = position//ncols+1
    column = (position%ncols)+1
    
    fig.add_trace(go.Scatter(x=Ndf['NDATE'], y=combined[col],mode='markers',
                            customdata = combined[col+' pct_change'],
                            hovertemplate="<br>".join(["Date: %{x}",
                                                       "Value: %{y}",
                                                       "Fraction Change: %{customdata}"
                                                       ])),      
                  row=row, col=column)
    position += 1
    
fig.update_layout(height=800, title_text="Metrics Positively Related with GDP",showlegend=False)
fig.update_xaxes(tickangle=45)

fig.show()

Inflation has surged significantly over the past year or two following COVID, with no signs of slowing down in the current data. Hopefully, the numbers for the coming years will show a decrease. The GDP growth rate has been highly volatile, with three years of negative growth in the past two decades. Notably, 2019 experienced a significant decline in GDP, followed by a period of strong growth in 2020, and another decline in 2021. Based on this data alone, trends for 2022 remain unpredictable.

The consumer price index for all items had a relatively low level of volatility. This will likely be a good measure to predict with linear regression. Let us do that later in the notebook.

Measure of Volatility¶

Lets create a correlation matrix to get a high level view of the relationship between metrics so that they can be analyzed furthur.

In [11]:
corrMatrix = Ndf.corr()
sns.heatmap(corrMatrix)
Out[11]:
<AxesSubplot:>
In [12]:
topcorr = corrMatrix['GDP PER CAPITA'].sort_values(ascending=False)[1:]
topcorr = topcorr.to_frame()

topcorr['GDP PER CAPITA'] = topcorr['GDP PER CAPITA'].map('{:,.3f}'.format)

fig = go.Figure(data=[go.Table(
    header=dict(values=['Metric','Correlation with GDP'],
                fill_color='paleturquoise',
                align='left'),
    cells=dict(values=[topcorr.index,topcorr.values],
               fill_color='lavender',
               align='left'))
])
fig.update_layout(title='GDP Correlations')
fig.show()

The table reports that Quarterly Real GDP, CPIALLITEMS, Median Household Income, and PPI-Const Mat. are very postively correlated while Mortgage interest monthly average, % share of working population, and corporate bond yield are negatively related to GDP.

GDP per capita¶

It's noted that one of the most comprehensive ways of measuring a country's economic health is its GDP. Above is a table showing the correlations of other metrics in the dataset with respect it.

Let us plot seperate visualizations to view the metrics positively and negatively correlated with GDP.

In [13]:
pct_change = Ndf.copy()
pct_change.set_index('NDATE',inplace=True)
pct_change = pct_change.pct_change()
for col in pct_change.columns:
    newcol = col+' pct_change'
    pct_change = pct_change.rename(columns={col:col+' pct_change'})
    pct_change[newcol] = pct_change[newcol].map('{:.0%}'.format)
    
combined = Ndf.join(pct_change,how='left',on='NDATE')
In [14]:
nrows = 3 
ncols = 2

cols = ['GDP PER CAPITA','CPIALLITEMS','MED HOUSEHOLD INCOME','PPI-CONST MAT.']
                                                     
fig = make_subplots(rows=nrows, cols=ncols,subplot_titles=cols,specs=[[{"colspan": 2}, None],[{}, {}],
           [{"colspan": 2}, None]],)

position = 0
for index,col in enumerate(cols):
    
    row = position//ncols+1
    column = (position%ncols)+1

    if row == 1 and column == 2:
        position += 1
        row = position//ncols+1
        column = (position%ncols)+1
    
    fig.add_trace(go.Scatter(x=combined['NDATE'], y=combined[col],mode='markers',
                            customdata = combined[col+' pct_change'],
                            hovertemplate="<br>".join(["Date: %{x}",
                                                       "Value: %{y}",
                                                       "Fraction Change: %{customdata}"
                                                       ])),      
                  row=row, col=column)
    position += 1
    
fig.update_layout(height=800, width=800, title_text="Metrics Positively Related with GDP",showlegend=False)
fig.update_xaxes(tickangle=45)

fig.show()

The graphs above display metrics positively correlated with GDP, including CPIALLITEMS, median household income, and PPI-CONST MAT. These metrics have generally trended upward, and you can hover over the data points to observe year-over-year changes. The relatively linear growth suggests these metrics represent more stable aspects of the economy. However, around 2020, all these indicators show notable volatility, almost certainly driven by the impact of COVID-19. Given this disruption, a linear model may not be suitable for predicting trends for 2023.

In [15]:
nrows = 3 
ncols = 2


cols = ['GDP PER CAPITA','CORP. BOND YIELD(%)','% SHARE OF WORKING POPULATION',
        'MORTGAGE INT. MONTHLY AVG(%)']
                                                     
fig = make_subplots(rows=nrows, cols=ncols,subplot_titles=cols,specs=[[{"colspan": 2}, None],[{}, {}],
           [{"colspan": 2}, None]],)

position = 0
for index,col in enumerate(cols):
    
    row = position//ncols+1
    column = (position%ncols)+1

    if row == 1 and column == 2:
        position += 1
        row = position//ncols+1
        column = (position%ncols)+1
    
    fig.add_trace(go.Scatter(x=combined['NDATE'], y=combined[col],mode='markers',
                            customdata = combined[col+' pct_change'],
                            hovertemplate="<br>".join(["Date: %{x}",
                                                       "Value: %{y}",
                                                       "Fraction Change: %{customdata}"
                                                       ])),      
                  row=row, col=column)
    position += 1
    
fig.update_layout(height=800, title_text="Metrics Negatively Related with GDP",showlegend=False)
fig.update_xaxes(tickangle=45)

fig.show()

The graphs above illustrate metrics negatively correlated with GDP. The percentage of the working population saw a slight increase before beginning a steady decline after 2008, with COVID appearing to slow this trend. Corporate bond yields have generally decreased over the past two decades but experienced a sharp rise around 2021–2022. Similarly, the monthly average mortgage rates closely followed this pattern, with declines over time and a significant uptick around 2022.

Highly Correlated Metrics¶

In [17]:
def createFig(df,x,columns=Ndf.columns.values,title='US Economic Metrics Over Time'):
    fig = px.line(df,x,y=columns,title = title)
    fig.update_xaxes(title='Date')
    fig.update_yaxes(title='Value')
    fig.update_layout(legend=dict(title='Metrics'))
    fig.show()

N2df = pd.DataFrame()
for column in df.columns:
    if df[column].dtype == 'int64' or df[column].dtype == 'float64':
        N2df[column] = df[column]/df[column].max()
N2df['DATE'] = df['DATE']
N2df['NDATE'] = N2df['DATE'].dt.strftime('%m-%Y')
N2df = N2df.groupby(by=['NDATE']).mean().reset_index()

createFig(N2df,'NDATE',['MED HOUSEHOLD INCOME','GDP PER CAPITA'],
         title = 'High Postive Correlated Metrics')

The Consumer Price Index of all items, Producer Price Index of construction materials, Median household income, and GDP per capita are all highly correlated, with a significant spike in the price of construction materials around 2020. It may be the case that this spike will cause a ripple effect in other metrics in the future.

Negatively Correlated Metrics¶

In [18]:
createFig(Ndf,'NDATE',['CONSUMER CONF INDEX','PPI-CONST MAT.'],
         title = 'High Negative Correlated Matrices')

A negative correlations was logged in the plot, however, there doesn't appear to be a solid relationship between the two. Consumer Confidence significantly dropped in 2009, very likely due to the housing crash in 2008.

In [19]:
createFig(N2df,'NDATE',['GDP PER CAPITA','% SHARE OF WORKING POPULATION'],
         title = 'GDP per capita and percent share of working populations')

GDP has still been going up significantly despite notable drops in the percent share of the working population

Predicting Consumer price index¶

The consumer price index for all items had a relatively low level of volatility. Let's take a look at it. This will likely be a good measure to predict with linear regression

In [20]:
# Predicting consumer price index
X = np.array(Ndf.index).reshape(-1,1)
y = Ndf['CPIALLITEMS']
model = LinearRegression()
model.fit(X,y)

fitted = pd.Series(model.predict(np.array([i for i in range(0,21)]).reshape(-1,1)))
fitted = fitted.to_frame()
fitted['NDATE'] = Ndf['NDATE']
fitted['type'] = ['fit']*len(fitted.index)

predictions = pd.Series(model.predict(np.array([i for i in range(20,26)]).reshape(-1,1)))
predictions = predictions.to_frame().reset_index()
predictions['index'] = predictions['index'] + 20
predictions['NDATE'] = [f"01-20{i}" for i in range(22,28)]
predictions['type'] = ['forecast']*len(predictions.index)
predictions.set_index('index',inplace=True)

combined = pd.concat([fitted,predictions])

fig = go.Figure()

fig.add_trace(go.Scatter(x=combined['NDATE'],y=combined[0].where(combined['type'] == 'fit'),
                    mode='lines',
                    name='Linear Fit',
                    hovertemplate="<br>".join(["Date: %{x}",
                                               "Fitted Price Index: %{y}",
                                                       ])))

fig.add_trace(go.Scatter(x=combined['NDATE'],y=combined[0].where(combined['type'] == 'forecast'),
                    mode='lines+text',
                    name='Forecast',text=combined[0].astype(int),textposition='top center',
                    hovertemplate="<br>".join(["Date: %{x}",
                                               "Forecasted Price Index: %{y}",
                                                       ])))

fig.add_trace(go.Scatter(x=Ndf['NDATE'], y=Ndf['CPIALLITEMS'],
                    mode='markers+text', name='CPIALLITEMS',text=Ndf['CPIALLITEMS'].astype(int),
                    textposition='top center',
                    hovertemplate="<br>".join(["Date: %{x}",
                                       "Actual Price Index: %{y}",
                                               ])))

fig.update_layout(title_text="Future Predictions for Consumer Price Index for All Items")
fig.update_xaxes(tickangle=45)

The green dots represent the actual values for CPIALLITEMS, while blue line is a linear fit of how it has progressed. In red are the forecasted values based on the linear fit for the next five years.

Predicting Inflation based on the data¶

In [21]:
df = pd.read_csv('US_Macroecon_data.csv')
df2 = df.copy()
df2.dropna()


# First we will isolate the features we want to use for our model
X = df[['UNRATE(%)', 'MORTGAGE INT. MONTHLY AVG(%)', 'MONTHLY HOME SUPPLY', 'GDP PER CAPITA', 'CONSUMER CONF INDEX']]
y = df['INFLATION(%)']

from sklearn.model_selection import train_test_split
# we will split the data intro training and testing data
X_train, X_test, y_train, y_test = train_test_split(X,y,test_size=0.2, random_state=1)

# we will use a random forest regressor model for this problem
from sklearn.ensemble import RandomForestRegressor
regressor = RandomForestRegressor(random_state=1)
regressor.fit(X_train, y_train)
predictions = regressor.predict(X_test)

# we will use the Mean Absolute Error metric to measure accuracy
from sklearn.metrics import mean_absolute_error
print(mean_absolute_error(y_test, predictions))

0.33628974646449

We can see that our model can predict the inflation rate with a mean absolute error of 0.33%. A very accurate prediction, which suggests the strong correlation between the metrics we used and the inflation rate.

In [22]:
# graphing the real inflation rates and the prediction results
real_inflation = list(y_test)
l = len(real_inflation)
x = range(l)
plt.plot(x, real_inflation, color='r', label='real inflation rates')
plt.plot(x, predictions, color='g', label='predicted inflation rates')
plt.xlabel('Index of data point')
plt.ylabel('Inflation Rate')
plt.title('Comparing predictions to reality on testing data')
plt.legend()
plt.show()

Since we found that these values are correlated with the interest rate let's see if we can use past values to predict future inflation rates.

In [23]:
# first we will add a new column where for each row it contains the inflation rate in the next month
future_inflation = [row['INFLATION(%)'] for index, row in df.iterrows()]
# dropping the first vlue
future_inflation.pop(0)
# dropping last row from data frame
df.drop(df.tail(1).index,inplace=True)
df['FutureInflation'] = future_inflation
df.head()
Out[23]:
DATE UNRATE(%) CONSUMER CONF INDEX PPI-CONST MAT. CPIALLITEMS INFLATION(%) MORTGAGE INT. MONTHLY AVG(%) MED HOUSEHOLD INCOME CORP. BOND YIELD(%) MONTHLY HOME SUPPLY % SHARE OF WORKING POPULATION GDP PER CAPITA QUARTERLY REAL GDP QUARTERLY GDP GROWTH RATE (%) CSUSHPISA FutureInflation
0 01-05-2022 3.6 106.4 352.857 123.322800 8.581511 5.2300 NaN 4.13 8.4 NaN 74737 19699.465 -0.144227 120.724 8.258629
1 01-04-2022 3.6 107.3 343.730 121.978170 8.258629 4.9825 NaN 3.76 8.4 NaN 74737 19699.465 -0.144227 121.813 8.542456
2 01-03-2022 3.6 107.2 345.852 121.301004 8.542456 4.1720 NaN 3.43 7.0 NaN 73289 19727.918 -0.395692 122.888 7.871064
3 01-02-2022 3.8 110.5 343.583 119.702806 7.871064 3.7625 NaN 3.25 6.0 NaN 73289 19727.918 -0.395692 123.831 7.479872
4 01-01-2022 4.0 113.8 345.742 118.619339 7.479872 3.4450 NaN 2.93 5.7 NaN 73289 19727.918 -0.395692 124.780 7.036403

Lets build the model.

In [24]:
X = df[['UNRATE(%)', 'MORTGAGE INT. MONTHLY AVG(%)', 'MONTHLY HOME SUPPLY', 'GDP PER CAPITA', 'CONSUMER CONF INDEX']]
y = df['FutureInflation']

from sklearn.model_selection import train_test_split
# we will split the data intro training and testing data
X_train, X_test, y_train, y_test = train_test_split(X,y,test_size=0.2, random_state=1)

# we will use a random forest regressor model for this problem
from sklearn.ensemble import RandomForestRegressor
regressor = RandomForestRegressor(random_state=1)
regressor.fit(X_train, y_train)
predictions = regressor.predict(X_test)

# we will use the Mean Absolute Error metric to measure accuracy
from sklearn.metrics import mean_absolute_error
print(mean_absolute_error(y_test, predictions))

0.34876464363479137

Our model can predict what the inflation rate will be next month with an error margin of 0.34%. That sounds good.

Lets plot the results

In [25]:
real_inflation = list(y_test)
l = len(real_inflation)
x = range(l)
plt.plot(x, real_inflation, color='r', label='real inflation rates')
plt.plot(x, predictions, color='g', label='predicted fututre inflation rates')
plt.xlabel('Index of data point')
plt.ylabel('Inflation Rate')
plt.title('Comparing predictions to reality on testing data')
plt.legend()
plt.show()

The prediction accuracy is at it's highest when we encounter high volatility in the inflation rate, similar to what we saw in our previous model.